An analysis of tax competition through VAR models
Raffaele Miniaci, Paolo Panteghini, Giulia Rivolta
Tax competition between countries has long been studied. On empirical grounds, most of the evidence is
based on a panel-approach and supports the tax competition hypothesis (e.g. Devereux et al., 2008), with
tax rates that are estimated to be, on average, strategic complements.
Unfortunately, for a given group of countries that compete with each other, the panel data approach
estimates only the effect of a change in the average taxation of competitors on the taxation of each country.
In doing so, the approach imposes restrictive conditions on tax interactions, mainly that: (i) each country
reacts to a weighted average of other countries' tax rates, (ii) the weights are exogenously given, and (iii) the
reaction to the tax competition variables is homogeneous across countries. Such restrictions do not fit with
the theoretical literature that, starting from Bucovetsky (1991), has stressed that the response functions may
be asymmetric so that a tax rate can be at the same time a strategic complement with respect to the tax rate
of a competing economy, and a strategic substitute with respect to a third country, where we have strategic
complements whenever the increase in one country tax rate leads another country to move its rate in the
same direction, and strategic substitutes otherwise (see Vrijburg and de Mooij, 2012). Therefore, given this
heterogeneity, the weighted average of the tax rates can hardly be considered a sufficient statistics to
describe tax competition.
We argue that these theoretical considerations are directly related to the appropriateness of the panel
models to empirically evaluate the presence of tax competition. The first aim of our paper is to unveil the
implications intrinsic in a dynamic panel model comparing it with a more flexible framework, i.e., a structural
vector autoregressive model. This analytical exercise shows that, in the panel framework, the heterogeneity
in the systematic reaction of tax rates to the contemporaneous and lagged tax rates of other countries is
solely determined by the choice of the exogenous weights, which also determine the impulse response
functions implied by the model.
The advantage of using panel models is that key parameters can be recovered even when the time series
dimension of the datasets is short, provided that there is sufficient cross sectional variability.
However, recent Bayesian econometric techniques allow to obtain robust inference in VAR models with
relatively few observations with respect to the number of variables. Therefore, we compare estimates
obtained from a panel model with those of a large Bayesian VAR model for a group of countries. Preliminary
results show that, in contrast to the findings in most of the previous empirical literature, the reactions to
exogenous shocks to other countries' tax rates are heterogeneous.
JEL codes: C54, H25, E62
Keywords: Tax competition, VAR models, Bayesian methods.